ambolt
About UsBlog
/
Contact

Services

AI Solutions DevelopmentCustom AI TrainingsAI Workspace Setup

Industries

ManufacturingEcommerceHR AgenciesForeign TradeAccounting FirmsMarketing AgenciesInsuranceHealth ServicesLegal & NotaryReal EstateFinancial ServicesLogistics
About UsBlog
/
Get in touch
StrategyProduct

How to measure whether AI improved your business (and not just whether it's used)

The few numbers that show whether the project is actually moving something.

Hands of a middle-aged person comparing two printed spreadsheets on an office desk, circling a figure with a pen, with a calculator and a mug to the side
Measuring impact means comparing the same process before and after, with the same numbers.

Measuring the impact of a project means answering a concrete question: after it goes live, does the business work better in the stretch that was chosen for improvement? The answer does not come from the usage dashboard or from spending on tools. It comes from two or three indicators defined in advance, compared with how that same process looked before.

In Argentina, the adoption of artificial intelligence is moving faster than the ability to organize it. A national survey by nadIA of 402 industrial and software SMEs found that 41.6% already use at least one AI technology, but among those that have adopted it, 57% do not formally measure its impact on business performance. In Latin America, Tech Pulse LATAM 2026 shows a similar gap: 67.9% report having AI in use and only 17.9% have it integrated into key processes. Adoption, on its own, does not guarantee a business result.

Usage, cost and impact: three distinct layers

When a team looks at "how the project is going," it often mixes three things:

  1. Usage. How many people open the tool, how many tasks go through it, how many times it is queried.
  2. Cost. Licenses, implementation time, maintenance, model consumption.
  3. Business impact. Cycle time, errors, conversion, hours freed up and reassigned, complaints, delays toward the customer.

Usage matters: a tool that nobody opens cannot improve anything. Cost matters so there are no surprises at the end of the month. Neither one stands in for impact. An assistant can show a lot of activity and still leave the same bottleneck in place; a cheap pilot can free up hours that later dissolve if nobody redirects that time.

That is why, when setting up the measurement, it helps to write the business indicator first and only then the usage one. The second explains whether the first has a chance of moving; it does not stand in for it.

Which indicators to choose (a few)

Two to four numbers are enough. More than that turns into a dashboard nobody looks at. Each one has to meet three conditions: it can be measured with data the company already has or can gather in a few weeks, it changes if the project works, and someone on the business side can understand it without a technical translation.

Some examples that often work in service and operations companies:

  • Cycle time of a request, an onboarding, a quote or a closing.
  • Error or rework rate in the same flow (corrections, returns, incomplete orders).
  • Team hours spent on that task per week, with a note on what they are reassigned to afterward.
  • Outward result: response time to the customer, conversion of an order, complaints tied to the process.

The choice comes from the problem that was meant to be solved, not from a generic list. If the pain was that quotes took days, the central indicator is that time; the rest is support. If the pain was errors in data entry, the central indicator is the correction rate.

The baseline is taken beforehand

Without a prior snapshot, any later improvement is only half-complete. It helps to measure the chosen process for two to four weeks before turning the solution on, using the same rule that will be used afterward. The same definition of a "closed case," the same time window, the same data source.

That snapshot does not have to be perfect. A simple spreadsheet, put together with the person who does the task every day, is enough to compare. What is not enough is remembering that "it used to be worse" with no numbers: selective memory favors the project and hides the exceptions.

When the project is already running and there was no baseline, a reference can still be built: a sample of older cases reconstructed, or a work group that continues the previous process for a short period. It is less clean, but better than not measuring at all.

Who defines the indicator

The indicator is chosen together by the person who lives the process and the person who answers for the result. Whoever leads sees the bottleneck and the cost of the error; whoever operates knows the exceptions and knows what can be counted without inventing extra work; finance or administration help translate hours and errors into a language that can be used to decide.

That same group agrees on the review frequency (every two weeks at the beginning is usually enough) and the minimum threshold to continue: what improvement, in what timeframe, justifies expanding the scope. Without that threshold, the pilot stretches on by inertia.

In work with companies in Latin America, the pattern that comes up most often is the reverse: first something is built, then a number is sought to justify what was done. Reversing the order shortens discussions and makes it clear, from day one, what is being bought.

How to read the first results

Within a few weeks, three frequent scenarios appear:

  • The business indicator moves and usage follows. It makes sense to expand carefully, adding the next stretch of the process.
  • Usage grows and the business indicator does not. There is adoption, but the bottleneck was somewhere else, or the tool solves a part that was not the one that hurt. In that case it helps to return to the diagnosis before adding features.
  • Neither usage nor impact. The problem may be one of design, of training, or of a process that was not yet ready to automate. Stopping or narrowing the scope is a valid decision.

The review exists to decide, not to dress up a report. Continue, adjust or cut are the three possible outcomes; measurement exists so that choice can be made with less opinion and more evidence.

Conclusion

Knowing whether a project is paying off is, in the end, a product habit: choose the stretch of the business that should move, write down how it looks today, build the minimum needed to move it and look at the same number again. When that habit takes hold, the conversation shifts from whether AI is being used to which process the next round should improve. That change of question is often the greatest impact of a first project that was measured well.

Frequently asked questions

How is the impact of an AI project measured in a company?

Two or three business indicators tied to the process that should improve are chosen, a baseline is recorded before implementation and the same process is compared afterward, with the same definition of a case. Tool usage is watched as a condition, not as the result.

What is the difference between measuring usage and measuring impact?

Usage counts activity (opens, tasks, queries). Impact counts whether the business improved in time, errors, conversion or another agreed result. Without impact, usage only shows that something is being turned on.

How many indicators are needed?

Two to four. Each one has to be understandable for the business, measurable with available data and sensitive to the change the project promises. A long dashboard usually ends up with no owner.

What should be done if the project is already running and there is no baseline?

Rebuild a reference with recent cases from the previous process, or measure in parallel a group that still works the old way. It is not ideal, but it makes it possible to decide on something more solid than the team's impression.


About Ambolt. Ambolt is a product agency with AI that works with companies in Latin America to choose what to build and to review, with business numbers, whether what was built is delivering real impact.

Ramiro MoyanoRamiro MoyanoPartner at Ambolt StudioPublished

Let's talk about your challenges.

We specialize in finding areas to optimize your company's operations, making them leaner, faster and more productive.

Talk to us →

Keep reading

All articles →